Object Detection Apparatus Using Head Estimation for Merging Results
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection techniques face challenges in accurately merging detection results from multiple detectors, particularly when objects are partially shielded or in varying postures, leading to erroneous positioning and reduced detection accuracy.
Innovation Solution
An object detection apparatus comprising multiple detection units, estimation units, and a determination unit to assess and merge detection results, using score correction and common site estimation to improve accuracy, especially by utilizing a head detector and entire body detector combination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple different detectors are used to detect objects in various postures and shielding conditions, then detection coverage is improved, but merging detection results becomes complex and error-prone
Solution Approach 1:
The patent introduces an estimation unit that acts as an intermediary between multiple detectors and the merging process. This unit estimates the position of a common site (head) based on detection results from different detectors (face detector, upper body detector, entire body detector), providing a unified reference point that simplifies the merging of detection results from multiple sources with different detection targets and coordinate systems.
2Productivity
If face detector and upper body detector results are simply merged, then processing speed is improved, but detection accuracy deteriorates when persons overlap
Solution Approach 1:
The estimation unit serves as a mediator that processes detection results from multiple detectors before final merging. It estimates the head position based on detection results from face, upper body, and entire body detectors, and uses this estimated position to determine which detector's result to prioritize when persons overlap, thereby maintaining both speed and accuracy.
Solution Approach 2:
The patent changes the parameter used for merging from simple coordinate overlay to estimated head position-based selection. By estimating the head position as an intermediate parameter and using it to determine the reliability of detection results, the system can accurately merge results even when persons overlap, without significantly increasing processing complexity.
3Device complexity
If face position is estimated from upper body detector result, then merging is simplified, but face position reliability decreases
Solution Approach 1:
The patent merges detection results from multiple detectors (face detector, upper body detector, entire body detector) to estimate the head position, rather than relying solely on the upper body detector. This combination of multiple detection sources improves the reliability of the estimated face position while still simplifying the overall merging process through the use of a common site estimation approach.
Data Source
AI summary
An object detection apparatus includes a first detection unit configured to detect a first portion of an object from an input image, a second detection unit configured to detect a second portion different from the first portion of the object, a first estimation unit configured to estimate a third portion of the object based on the first portion, a second estimation unit configured to estimate a third portion of the object based on the second portion, a determination unit configured to determine whether the third portions, which have been respectively estimated by the first and second estimation units, match each other, and an output unit configured to output, if the third portions match each other, a detection result of the object based on at least one of a detection result of the first or second detection unit and an estimation result of the first or second estimation unit.